Data-Driven Approach to Accommodating Multiple Simultaneous Sensor Faults in Variable-Gain PID Systems

Abstract

This paper is concerned with the problem of data-driven fault-tolerant control for multiple simultaneous sensor drift faults in variable-gain digital PID systems with very large time constants and long dead time, which are exceedingly common characteristics of process control systems. No existing data-driven residual generation method can allow building (with low computational costs) residual variables independent of the state of these systems, and meanwhile guarantee that each of the sensor faults is mapped uniquely and entirely onto the associated residual variable. To solve the aforementioned technical difficulty, a novel residual generation technique is devised via the dead time as well as the coefficients and state of the variable-gain PID controller. On this basis, a methodology is developed for the purpose of the full-decoupling estimation of several sensor malfunctions from the residual variables. Finally, a resulting data-driven approach to compensate for the aforesaid faults is applied to a dual-chamber electric heating furnace (which is a typical process plant), so that the effectiveness and advantages of the proposed methods are verified by experiment.